[UX] Reduce startup log noise (#50590)

Signed-off-by: mgoin <mgoin64@gmail.com>
Co-authored-by: OpenAI Codex <noreply@openai.com>
This commit is contained in:
Michael Goin
2026-07-31 13:04:34 -07:00
committed by GitHub
co-authored by OpenAI Codex
parent e8b358b4c0
commit 90329913e9
9 changed files with 36 additions and 50 deletions
@@ -107,7 +107,7 @@ class CustomAllreduce:
if not custom_ar:
# disable because of missing custom allreduce library
# e.g. in a non-GPU environment
logger.info(
logger.info_once(
"Custom allreduce is disabled because "
"of missing custom allreduce library"
)
@@ -130,7 +130,7 @@ class CustomAllreduce:
return
if world_size not in CustomAllreduce._SUPPORTED_WORLD_SIZES:
logger.warning(
logger.warning_once(
"Custom allreduce is disabled due to an unsupported world"
" size: %d. Supported world sizes: %s. To silence this "
"warning, specify disable_custom_all_reduce=True explicitly.",
@@ -328,7 +328,7 @@ class CustomAllreduce:
def register_graph_buffers(self):
handle, offset = ops.get_graph_buffer_ipc_meta(self._ptr)
logger.info("Registering %d cuda graph addresses", len(offset))
logger.debug("Registering %d cuda graph addresses", len(offset))
# We cannot directly use `dist.all_gather_object` here
# because it is incompatible with `gloo` backend under inference mode.
# see https://github.com/pytorch/pytorch/issues/126032 for details.
@@ -121,7 +121,7 @@ def _resolve_fi_ar_backend() -> tuple[str, bool]:
"""
backend = envs.VLLM_FLASHINFER_ALLREDUCE_BACKEND
if backend != "auto":
logger.info_once(f"Using flashinfer allreduce backend: {backend}")
logger.debug_once("Using flashinfer allreduce backend: %s", backend)
return backend, False
# Default to mnnvl for both single- and multi-node setups. The mnnvl
@@ -134,7 +134,7 @@ def _resolve_fi_ar_backend() -> tuple[str, bool]:
backend = "mnnvl"
allow_trtllm_fallback = get_node_count() == 1
logger.info_once(f"Auto-selected flashinfer allreduce backend: {backend}")
logger.debug_once("Auto-selected flashinfer allreduce backend: %s", backend)
return backend, allow_trtllm_fallback